second brain

Stop feeding your entire knowledge base into context windows or cloud-hosted notebooks. This skill implements a local-first retrieval system that processes your private Markdown vault using local LLMs. It solves the privacy problem by ensuring sensitive project notes and technical documentation stay on your disk, only sending specific, retrieved context packs to a local model for synthesis.

What it does

• Vault Indexing generates a local vector store from your Markdown notes for efficient retrieval.
• Privacy Enforcement ensures note content stays local, preventing large-scale data leakage to cloud providers.
• Synthesis Engine transforms raw notes into structured project briefs, research summaries, and technical proposals.
• Decision Tracking records architecture decisions directly back to your vault as formatted Markdown files.
• Risk Assessment automatically appends technical and timeline risks to every generated document based on vault hooks.

How it works

1. Index the vault using make index whenever your local Markdown files are updated.
2. Query the system using specific commands like make research or make brief with a targeted subject string.
3. Verify the output, which combines retrieved context with mandatory risk and assumption sections.
4. Persist decisions by running make decision, which saves a timestamped record to the vault.

Frameworks & tools

Built with Python 3.10+, GNU Make, and Ollama. It requires the llama3 model and utilizes Chroma for local vector storage of Markdown files.

Why this beats prompting it yourself

Manual prompting requires copy-pasting notes, which is slow and risks leaking data to the model provider. This skill automates the RAG pipeline locally, ensuring consistent formatting and mandatory risk analysis that single prompts often skip.

Use cases

• Draft project briefs using scattered meeting notes and technical specs.
• Generate technical proposals from private internal documentation.
• Synthesize research findings across dozens of disparate Markdown files.
• Log architectural decisions with automated risk and assumption tagging.

Known limitations

Requires a local Ollama instance running llama3. The vault must contain Markdown files to function. It does not support non-markdown file formats or remote vector databases.